linux/drivers/block/brd.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Ram backed block device driver.
   4 *
   5 * Copyright (C) 2007 Nick Piggin
   6 * Copyright (C) 2007 Novell Inc.
   7 *
   8 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
   9 * of their respective owners.
  10 */
  11
  12#include <linux/init.h>
  13#include <linux/initrd.h>
  14#include <linux/module.h>
  15#include <linux/moduleparam.h>
  16#include <linux/major.h>
  17#include <linux/blkdev.h>
  18#include <linux/bio.h>
  19#include <linux/highmem.h>
  20#include <linux/mutex.h>
  21#include <linux/radix-tree.h>
  22#include <linux/fs.h>
  23#include <linux/slab.h>
  24#include <linux/backing-dev.h>
  25
  26#include <linux/uaccess.h>
  27
  28#define PAGE_SECTORS_SHIFT      (PAGE_SHIFT - SECTOR_SHIFT)
  29#define PAGE_SECTORS            (1 << PAGE_SECTORS_SHIFT)
  30
  31/*
  32 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
  33 * the pages containing the block device's contents. A brd page's ->index is
  34 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
  35 * with, the kernel's pagecache or buffer cache (which sit above our block
  36 * device).
  37 */
  38struct brd_device {
  39        int             brd_number;
  40
  41        struct request_queue    *brd_queue;
  42        struct gendisk          *brd_disk;
  43        struct list_head        brd_list;
  44
  45        /*
  46         * Backing store of pages and lock to protect it. This is the contents
  47         * of the block device.
  48         */
  49        spinlock_t              brd_lock;
  50        struct radix_tree_root  brd_pages;
  51};
  52
  53/*
  54 * Look up and return a brd's page for a given sector.
  55 */
  56static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
  57{
  58        pgoff_t idx;
  59        struct page *page;
  60
  61        /*
  62         * The page lifetime is protected by the fact that we have opened the
  63         * device node -- brd pages will never be deleted under us, so we
  64         * don't need any further locking or refcounting.
  65         *
  66         * This is strictly true for the radix-tree nodes as well (ie. we
  67         * don't actually need the rcu_read_lock()), however that is not a
  68         * documented feature of the radix-tree API so it is better to be
  69         * safe here (we don't have total exclusion from radix tree updates
  70         * here, only deletes).
  71         */
  72        rcu_read_lock();
  73        idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
  74        page = radix_tree_lookup(&brd->brd_pages, idx);
  75        rcu_read_unlock();
  76
  77        BUG_ON(page && page->index != idx);
  78
  79        return page;
  80}
  81
  82/*
  83 * Look up and return a brd's page for a given sector.
  84 * If one does not exist, allocate an empty page, and insert that. Then
  85 * return it.
  86 */
  87static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
  88{
  89        pgoff_t idx;
  90        struct page *page;
  91        gfp_t gfp_flags;
  92
  93        page = brd_lookup_page(brd, sector);
  94        if (page)
  95                return page;
  96
  97        /*
  98         * Must use NOIO because we don't want to recurse back into the
  99         * block or filesystem layers from page reclaim.
 100         */
 101        gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM;
 102        page = alloc_page(gfp_flags);
 103        if (!page)
 104                return NULL;
 105
 106        if (radix_tree_preload(GFP_NOIO)) {
 107                __free_page(page);
 108                return NULL;
 109        }
 110
 111        spin_lock(&brd->brd_lock);
 112        idx = sector >> PAGE_SECTORS_SHIFT;
 113        page->index = idx;
 114        if (radix_tree_insert(&brd->brd_pages, idx, page)) {
 115                __free_page(page);
 116                page = radix_tree_lookup(&brd->brd_pages, idx);
 117                BUG_ON(!page);
 118                BUG_ON(page->index != idx);
 119        }
 120        spin_unlock(&brd->brd_lock);
 121
 122        radix_tree_preload_end();
 123
 124        return page;
 125}
 126
 127/*
 128 * Free all backing store pages and radix tree. This must only be called when
 129 * there are no other users of the device.
 130 */
 131#define FREE_BATCH 16
 132static void brd_free_pages(struct brd_device *brd)
 133{
 134        unsigned long pos = 0;
 135        struct page *pages[FREE_BATCH];
 136        int nr_pages;
 137
 138        do {
 139                int i;
 140
 141                nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
 142                                (void **)pages, pos, FREE_BATCH);
 143
 144                for (i = 0; i < nr_pages; i++) {
 145                        void *ret;
 146
 147                        BUG_ON(pages[i]->index < pos);
 148                        pos = pages[i]->index;
 149                        ret = radix_tree_delete(&brd->brd_pages, pos);
 150                        BUG_ON(!ret || ret != pages[i]);
 151                        __free_page(pages[i]);
 152                }
 153
 154                pos++;
 155
 156                /*
 157                 * It takes 3.4 seconds to remove 80GiB ramdisk.
 158                 * So, we need cond_resched to avoid stalling the CPU.
 159                 */
 160                cond_resched();
 161
 162                /*
 163                 * This assumes radix_tree_gang_lookup always returns as
 164                 * many pages as possible. If the radix-tree code changes,
 165                 * so will this have to.
 166                 */
 167        } while (nr_pages == FREE_BATCH);
 168}
 169
 170/*
 171 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
 172 */
 173static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
 174{
 175        unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
 176        size_t copy;
 177
 178        copy = min_t(size_t, n, PAGE_SIZE - offset);
 179        if (!brd_insert_page(brd, sector))
 180                return -ENOSPC;
 181        if (copy < n) {
 182                sector += copy >> SECTOR_SHIFT;
 183                if (!brd_insert_page(brd, sector))
 184                        return -ENOSPC;
 185        }
 186        return 0;
 187}
 188
 189/*
 190 * Copy n bytes from src to the brd starting at sector. Does not sleep.
 191 */
 192static void copy_to_brd(struct brd_device *brd, const void *src,
 193                        sector_t sector, size_t n)
 194{
 195        struct page *page;
 196        void *dst;
 197        unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
 198        size_t copy;
 199
 200        copy = min_t(size_t, n, PAGE_SIZE - offset);
 201        page = brd_lookup_page(brd, sector);
 202        BUG_ON(!page);
 203
 204        dst = kmap_atomic(page);
 205        memcpy(dst + offset, src, copy);
 206        kunmap_atomic(dst);
 207
 208        if (copy < n) {
 209                src += copy;
 210                sector += copy >> SECTOR_SHIFT;
 211                copy = n - copy;
 212                page = brd_lookup_page(brd, sector);
 213                BUG_ON(!page);
 214
 215                dst = kmap_atomic(page);
 216                memcpy(dst, src, copy);
 217                kunmap_atomic(dst);
 218        }
 219}
 220
 221/*
 222 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
 223 */
 224static void copy_from_brd(void *dst, struct brd_device *brd,
 225                        sector_t sector, size_t n)
 226{
 227        struct page *page;
 228        void *src;
 229        unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
 230        size_t copy;
 231
 232        copy = min_t(size_t, n, PAGE_SIZE - offset);
 233        page = brd_lookup_page(brd, sector);
 234        if (page) {
 235                src = kmap_atomic(page);
 236                memcpy(dst, src + offset, copy);
 237                kunmap_atomic(src);
 238        } else
 239                memset(dst, 0, copy);
 240
 241        if (copy < n) {
 242                dst += copy;
 243                sector += copy >> SECTOR_SHIFT;
 244                copy = n - copy;
 245                page = brd_lookup_page(brd, sector);
 246                if (page) {
 247                        src = kmap_atomic(page);
 248                        memcpy(dst, src, copy);
 249                        kunmap_atomic(src);
 250                } else
 251                        memset(dst, 0, copy);
 252        }
 253}
 254
 255/*
 256 * Process a single bvec of a bio.
 257 */
 258static int brd_do_bvec(struct brd_device *brd, struct page *page,
 259                        unsigned int len, unsigned int off, unsigned int op,
 260                        sector_t sector)
 261{
 262        void *mem;
 263        int err = 0;
 264
 265        if (op_is_write(op)) {
 266                err = copy_to_brd_setup(brd, sector, len);
 267                if (err)
 268                        goto out;
 269        }
 270
 271        mem = kmap_atomic(page);
 272        if (!op_is_write(op)) {
 273                copy_from_brd(mem + off, brd, sector, len);
 274                flush_dcache_page(page);
 275        } else {
 276                flush_dcache_page(page);
 277                copy_to_brd(brd, mem + off, sector, len);
 278        }
 279        kunmap_atomic(mem);
 280
 281out:
 282        return err;
 283}
 284
 285static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio)
 286{
 287        struct brd_device *brd = bio->bi_disk->private_data;
 288        struct bio_vec bvec;
 289        sector_t sector;
 290        struct bvec_iter iter;
 291
 292        sector = bio->bi_iter.bi_sector;
 293        if (bio_end_sector(bio) > get_capacity(bio->bi_disk))
 294                goto io_error;
 295
 296        bio_for_each_segment(bvec, bio, iter) {
 297                unsigned int len = bvec.bv_len;
 298                int err;
 299
 300                /* Don't support un-aligned buffer */
 301                WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
 302                                (len & (SECTOR_SIZE - 1)));
 303
 304                err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
 305                                  bio_op(bio), sector);
 306                if (err)
 307                        goto io_error;
 308                sector += len >> SECTOR_SHIFT;
 309        }
 310
 311        bio_endio(bio);
 312        return BLK_QC_T_NONE;
 313io_error:
 314        bio_io_error(bio);
 315        return BLK_QC_T_NONE;
 316}
 317
 318static int brd_rw_page(struct block_device *bdev, sector_t sector,
 319                       struct page *page, unsigned int op)
 320{
 321        struct brd_device *brd = bdev->bd_disk->private_data;
 322        int err;
 323
 324        if (PageTransHuge(page))
 325                return -ENOTSUPP;
 326        err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
 327        page_endio(page, op_is_write(op), err);
 328        return err;
 329}
 330
 331static const struct block_device_operations brd_fops = {
 332        .owner =                THIS_MODULE,
 333        .rw_page =              brd_rw_page,
 334};
 335
 336/*
 337 * And now the modules code and kernel interface.
 338 */
 339static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
 340module_param(rd_nr, int, 0444);
 341MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
 342
 343unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
 344module_param(rd_size, ulong, 0444);
 345MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
 346
 347static int max_part = 1;
 348module_param(max_part, int, 0444);
 349MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
 350
 351MODULE_LICENSE("GPL");
 352MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
 353MODULE_ALIAS("rd");
 354
 355#ifndef MODULE
 356/* Legacy boot options - nonmodular */
 357static int __init ramdisk_size(char *str)
 358{
 359        rd_size = simple_strtol(str, NULL, 0);
 360        return 1;
 361}
 362__setup("ramdisk_size=", ramdisk_size);
 363#endif
 364
 365/*
 366 * The device scheme is derived from loop.c. Keep them in synch where possible
 367 * (should share code eventually).
 368 */
 369static LIST_HEAD(brd_devices);
 370static DEFINE_MUTEX(brd_devices_mutex);
 371
 372static struct brd_device *brd_alloc(int i)
 373{
 374        struct brd_device *brd;
 375        struct gendisk *disk;
 376
 377        brd = kzalloc(sizeof(*brd), GFP_KERNEL);
 378        if (!brd)
 379                goto out;
 380        brd->brd_number         = i;
 381        spin_lock_init(&brd->brd_lock);
 382        INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
 383
 384        brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
 385        if (!brd->brd_queue)
 386                goto out_free_dev;
 387
 388        blk_queue_make_request(brd->brd_queue, brd_make_request);
 389
 390        /* This is so fdisk will align partitions on 4k, because of
 391         * direct_access API needing 4k alignment, returning a PFN
 392         * (This is only a problem on very small devices <= 4M,
 393         *  otherwise fdisk will align on 1M. Regardless this call
 394         *  is harmless)
 395         */
 396        blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
 397        disk = brd->brd_disk = alloc_disk(max_part);
 398        if (!disk)
 399                goto out_free_queue;
 400        disk->major             = RAMDISK_MAJOR;
 401        disk->first_minor       = i * max_part;
 402        disk->fops              = &brd_fops;
 403        disk->private_data      = brd;
 404        disk->flags             = GENHD_FL_EXT_DEVT;
 405        sprintf(disk->disk_name, "ram%d", i);
 406        set_capacity(disk, rd_size * 2);
 407        brd->brd_queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO;
 408
 409        /* Tell the block layer that this is not a rotational device */
 410        blk_queue_flag_set(QUEUE_FLAG_NONROT, brd->brd_queue);
 411        blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, brd->brd_queue);
 412
 413        return brd;
 414
 415out_free_queue:
 416        blk_cleanup_queue(brd->brd_queue);
 417out_free_dev:
 418        kfree(brd);
 419out:
 420        return NULL;
 421}
 422
 423static void brd_free(struct brd_device *brd)
 424{
 425        put_disk(brd->brd_disk);
 426        blk_cleanup_queue(brd->brd_queue);
 427        brd_free_pages(brd);
 428        kfree(brd);
 429}
 430
 431static struct brd_device *brd_init_one(int i, bool *new)
 432{
 433        struct brd_device *brd;
 434
 435        *new = false;
 436        list_for_each_entry(brd, &brd_devices, brd_list) {
 437                if (brd->brd_number == i)
 438                        goto out;
 439        }
 440
 441        brd = brd_alloc(i);
 442        if (brd) {
 443                brd->brd_disk->queue = brd->brd_queue;
 444                add_disk(brd->brd_disk);
 445                list_add_tail(&brd->brd_list, &brd_devices);
 446        }
 447        *new = true;
 448out:
 449        return brd;
 450}
 451
 452static void brd_del_one(struct brd_device *brd)
 453{
 454        list_del(&brd->brd_list);
 455        del_gendisk(brd->brd_disk);
 456        brd_free(brd);
 457}
 458
 459static struct kobject *brd_probe(dev_t dev, int *part, void *data)
 460{
 461        struct brd_device *brd;
 462        struct kobject *kobj;
 463        bool new;
 464
 465        mutex_lock(&brd_devices_mutex);
 466        brd = brd_init_one(MINOR(dev) / max_part, &new);
 467        kobj = brd ? get_disk_and_module(brd->brd_disk) : NULL;
 468        mutex_unlock(&brd_devices_mutex);
 469
 470        if (new)
 471                *part = 0;
 472
 473        return kobj;
 474}
 475
 476static int __init brd_init(void)
 477{
 478        struct brd_device *brd, *next;
 479        int i;
 480
 481        /*
 482         * brd module now has a feature to instantiate underlying device
 483         * structure on-demand, provided that there is an access dev node.
 484         *
 485         * (1) if rd_nr is specified, create that many upfront. else
 486         *     it defaults to CONFIG_BLK_DEV_RAM_COUNT
 487         * (2) User can further extend brd devices by create dev node themselves
 488         *     and have kernel automatically instantiate actual device
 489         *     on-demand. Example:
 490         *              mknod /path/devnod_name b 1 X   # 1 is the rd major
 491         *              fdisk -l /path/devnod_name
 492         *      If (X / max_part) was not already created it will be created
 493         *      dynamically.
 494         */
 495
 496        if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
 497                return -EIO;
 498
 499        if (unlikely(!max_part))
 500                max_part = 1;
 501
 502        for (i = 0; i < rd_nr; i++) {
 503                brd = brd_alloc(i);
 504                if (!brd)
 505                        goto out_free;
 506                list_add_tail(&brd->brd_list, &brd_devices);
 507        }
 508
 509        /* point of no return */
 510
 511        list_for_each_entry(brd, &brd_devices, brd_list) {
 512                /*
 513                 * associate with queue just before adding disk for
 514                 * avoiding to mess up failure path
 515                 */
 516                brd->brd_disk->queue = brd->brd_queue;
 517                add_disk(brd->brd_disk);
 518        }
 519
 520        blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
 521                                  THIS_MODULE, brd_probe, NULL, NULL);
 522
 523        pr_info("brd: module loaded\n");
 524        return 0;
 525
 526out_free:
 527        list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
 528                list_del(&brd->brd_list);
 529                brd_free(brd);
 530        }
 531        unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 532
 533        pr_info("brd: module NOT loaded !!!\n");
 534        return -ENOMEM;
 535}
 536
 537static void __exit brd_exit(void)
 538{
 539        struct brd_device *brd, *next;
 540
 541        list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
 542                brd_del_one(brd);
 543
 544        blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
 545        unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 546
 547        pr_info("brd: module unloaded\n");
 548}
 549
 550module_init(brd_init);
 551module_exit(brd_exit);
 552
 553